[BUG]image margin with no DAPI detected huge amounts of cells when croped to predict using cellpose2

[rmd13]

Describe the bug
I cropped the images into small tiles. Some tiles are close to the corner and have no DAPI but only extreme background patches.
For this tile, I found that it predict hundreds of "cells".
For some other marginal tiles, once there is even a single DAPI spot present, it predict correctly of that DAPI.
So the problem only take place when there is absolute no DAPI in the tiles.

Is there a way to solve this problem?